A lossless copyright authentication scheme based on Bessel–Fourier moment and extreme learning machine in curvature-feature domain

Abstract

To overcome some drawbacks existing in current zero-watermarking methods, a lossless copyright authentication scheme is proposed in this paper. This scheme designs a multiple zero-watermarking algorithm based on Bessel–Fourier moment and extreme learning machine (ELM) in curvature-feature domain, develops a method for image feature enhancement and noise suppression in curvature-feature domain, and presents a simple algorithm which uses Bessel–Fourier moment phase to estimate the rotation angle of the rotation-attacked image. The experimental results, involving five types of images, indicate the proposed scheme has better overall performance compared to other five current methods, especially in the aspects of resisting high ratio cropping and large angle rotation attacks. Finally, some related factors including phase and magnitude components, feature vector dimension and ELM optimization are considered in the algorithm performance evaluation.